Ambient Ionization and FAIMS Mass Spectrometry for EnhancedImaging of Multiply Charged Molecular Ions in Biological TissuesClara L. Feider, Natalia Elizondo, and Livia S. Eberlin*

Department of Chemistry, The University of Texas at Austin, Austin, Texas 78712, United States

*S Supporting Information

ABSTRACT: Ambient ionization mass spectrometry imaging(MSI) has been increasingly used to investigate the moleculardistribution of biological tissue samples. Here, we report theintegration and optimization of desorption electrosprayionization (DESI) and liquid-microjunction surface samplingprobe (LMJ-SSP) with a chip-based high-field asymmetricwaveform ion mobility spectrometry (FAIMS) device to imagemetabolites, lipids, and proteins in biological tissue samples.Optimized FAIMS parameters for specific molecular classesenabled semitargeted detection of multiply charged molecularspecies at enhanced signal-to-noise ratios (S/N), improvedvisualization of spatial distributions, and, most importantly,allowed detection of species which were unseen by ambientionization MSI alone. Under static DESI-FAIMS conditions selected for transmission of doubly charged cardiolipins (CL), forexample, detection of 71 different CL species was achieved in rat brain, 23 of which were not observed by DESI alone. DiagnosticCL were imaged in a human thyroid tumor sample with reduced interference of isobaric species. LMJ-SSP-FAIMS enableddetection of 84 multiply charged protein ions in rat brain tissue, 66 of which were exclusive to this approach. Spatial visualizationof proteins in substructures of rat brain, and in human ovarian cancerous, necrotic, and normal tissues was achieved. Our resultsindicate that integration of FAIMS with ambient ionization MS allows improved detection and imaging of selected molecularspecies. We show that this methodology is valuable in biomedical applications of MSI for detection of multiply charged lipids andproteins from biological tissues.

Mass spectrometry imaging (MSI) provides the out-standing ability of probing the spatial distribution of

molecules in a sample surface with high specificity andsensitivity.1−3 In particular, ambient ionization MSI techniqueshave revolutionized the means by which spatial and molecularinformation is obtained from biological samples by enabling insitu, real time analysis of tissue samples with minimalpretreatment.4,5 Desorption electrospray ionization (DESI) isthe most commonly used solvent-based ambient ionization MStechnique to image and characterize lipids and metabolites inbiological tissue samples.6,7 DESI has been increasingly appliedfor cancer diagnosis with the perspective of clinical use.8 Inaddition to DESI, other solvent-based ambient ionizationtechniques have been used to analyze biological tissue samples,including liquid microjunction surface sampling probe (LMJ-SSP)11 and nanoDESI.9,10

Despite providing a wealth of chemical and spatialinformation, inherent challenges of sample complexity usingdirect analysis by MSI have prevented comprehensive detectionand characterization of molecular species.11,12 Naturallyoccurring lipids, for example, present enormous diversity ofmolecular structures and are observed over a relatively narrowmass-to-charge (m/z) range as molecular ions. Isobaricinterferences in the mass spectrum complicate tandem MSanalysis for structural characterization and obscures visual-

ization of an ion’s spatial distribution,13 an issue that has beenpreviously addressed through MS/MS imaging.14 Lipidspresent in high concentration in biological tissues may alsosuppress the detection of other lipids at lower abundances.15

Protein imaging directly from biological samples is also anongoing analytical challenge for ambient ionization techniques,although recent progress has been made.16 New approachesthat integrate several analytical strategies are needed forimproved imaging and characterization of molecules withinbiological tissue samples.17,18

Ion mobility has been increasingly applied to overcomeissues in complex sample analysis by MS.19−21 In particular,high-field asymmetric waveform ion mobility spectrometry(FAIMS), or differential mobility separation, separates gasphase ions at atmospheric pressure on the basis of differences intheir mobilities in electric fields prior to MS analysis.22 A high-frequency, asymmetric waveform is employed by alternatinglow and high electric fields perpendicular to the path throughwhich the ions travel.23 This waveform, called the dispersionfield (DF), causes the ions to be displaced from their initial

Received: July 21, 2016Accepted: October 26, 2016Published: October 26, 2016

Article

pubs.acs.org/ac

© 2016 American Chemical Society 11533 DOI: 10.1021/acs.analchem.6b02798Anal. Chem. 2016, 88, 11533−11541

trajectories, collide with the electrode walls, and be dissipated.However, a second, smaller DC voltage can be applied betweenthe two electrode plates, creating a low-energy electric field thatis superimposed on the DF. This compensation field (CF)counteracts the ion drift caused by the DF, allowing ions withspecific mobilities to be transmitted for mass analysis. The high-speed gas phase separation provided by FAIMS has been shownto reduce chemical noise and to improve signal-to-noise ratios(S/N), sensitivity, and dynamic range.22 By optimizing the DFand CF, semiselective transmission of subsets of ions or classesof molecules can be achieved. The filtering capability of FAIMSallows only subsets of ions to enter and fill the mass analyzer,resulting in higher sensitivity in their detection. Several groupshave explored the use of FAIMS, as well as other ion mobilitytechniques, for MSI applications.24−28 A lab-built DMS cellintegrated with a DESI platform was used to imageglycerophosphocholines (PC) in a mouse brain tissue, resultingin decreased chemical noise, as well as improved S/N andimage contrast.29 More recently, liquid extraction surfaceanalysis (LESA) was integrated with the FAIMS device weapplied in our study to image proteins in mouse brain and livertissue samples.30,31 Using the integrated LESA-FAIMSapproach, 26 protein species in mouse brain and 29 proteinspecies in mouse liver undetectable through LESA alone wereobserved.31 Here, we report an analytical approach couplingDESI or LMJ-SSP with FAIMS to image and characterize lipidsand proteins from biological tissue sections.In our study, we integrated and optimized a chip-based, high-

speed ultraFAIMS device with DESI-MS/LMJ-SSP-MS and ahigh mass resolution mass spectrometer to image andcharacterize singly charged metabolites, singly- and doublycharged glycerophospholipids (GP) and glycosphingolipids,and multiply charged proteins in rat brain, human thyroid, andhuman ovarian cancer tissues. 2D-FAIMS sweep experimentswere performed in tandem with DESI-MSI using a spot-by-spotapproach to explore separation voltages for specific molecularclasses of lipids and metabolites in a rat brain tissue section.Static, optimized FAIMS conditions were then used for doublycharged CL and gangliosides with no increase in analysis time.Multiply charged protein ions were imaged in biological tissuesusing LMJ-SSP-FAIMS with enhanced analytical performance.Using this approach, we show the first example of globalprotein imaging in human cancerous tissue by ambientionization MSI. Our results indicate that integration ofFAIMS with DESI or LMJ-SSP is valuable for imaging selectedmolecular classes in biological tissues.

■ EXPERIMENTAL SECTIONChemicals. 18:1 glycerophosphoinositol (PI), 18:1 glycer-

ophosphoserine (PS), 18:1 glycerophosphoethanolamine (PE),18:1 glycerophosphoglycerol (PG), 18:1 monoacylglycerophos-phate (PA), 18:1 cardiolipin (CL), and a total gangliosideextract were purchased from Avanti Polar Lipids (Alabaster,AL). Fatty acids (FA) including oleic acid, lignoceric acid, andarachidonic acid, as well as metabolites n-acetylaspartic acid and2-hydroxyglutarate were purchased from Sigma-Aldrich (St.Louis, MO). Standards were dissolved in chloroform/methanol(CHCl3/MeOH) 1:1 (v/v) at a concentration of 10 μg/mL. Aubiquitin standard (Sigma-Aldrich, St. Louis, MO) wasdissolved in acetonitrile/water (ACN/H2O) 60:40 (v/v) with0.1% formic acid at 1 μg/mL. Direct electrospray ionization(ESI) infusion of the standards was performed at a flow rate of5 μL/min.

Tissue Samples. Rat brain samples were obtained fromBioreclamationIVT (Hicksville, NY). Banked frozen humantissue samples including human thyroid oncocytic tumor,ovarian cancer, and normal ovarian tissues were obtained fromCooperative Human Tissue Network (CHTN) under approvedIRB protocol. Samples were stored in a −80 °C freezer untilsectioned. Tissue samples were sectioned at a thickness of 16μm using a CryoStar NX50 cryostat (Thermo Scientific, SanJose, CA). After sectioning, the glass slides were stored in a−80 °C freezer. Prior to MSI, the glass slides were dried for∼15 min. For protein analysis, sections were washed in ethanolfor 10 s, followed by a wash in chloroform for 10 s to removeexcess lipids.

FAIMS. An ultraFAIMS device (Owlstone Ltd., Cambridge,UK) using an ND chip (Owlstone), was used for allexperiments performed. The ND chip has a trench length of97.0 μm, a gap width of 101.51 μm, and a chip thickness of 700μm. The chip-region temperature was set to 90 °C. At thistemperature, the DF can be set to any value from 0 to 280.98Townsends (Td) and the CF from −14.84 to 14.84 Td.

DESI-MSI. A 2D Omni Spray (Prosolia Inc., Indianapolis,IN) coupled to a Q Exactive mass spectrometer (ThermoScientific, San Jose, CA) was used for tissue profiling andimaging. An unheated extended transfer tube of 5 in. with aninner diameter of 0.067 in. was lab-built to fit the FAIMSplatform. DESI-MSI was performed in the negative ion modefrom m/z 100−1500, using the Q Exactive mass spectrometerwhich allows for high mass accuracy (<5 ppm mass error) andhigh mass resolution (70 000 resolving power at m/z 200)measurements. The spatial resolution of the imaging experi-ments was 200 μm. The histologically compatible solventsystem dimethylformamide/acetonitrile (DMF/ACN) 1:1 (v/v) was used for analysis at a flow rate of 1.2 μL/min with 5 kVapplied to the solvent.32 The N2 pressure was set to 180 psi.The capillary temperature of the mass spectrometer was set to300 °C.

LMJ-SSP MSI. A LMJ-SSP system, the Flowprobe (ProsoliaInc. Indianapolis, IN), coupled to a Q Exactive massspectrometer (Thermo Scientific, San Jose, CA) was used fortissue profiling and imaging. LMJ-SSP-MSI was performed inthe positive ion mode from m/z 250−2000, using the hybridQuadrupole-Orbitrap mass spectrometer. The spatial resolutionof the imaging experiments was ∼630 μm. The solvent systemof acetonitrile/water (ACN/H2O) 60:40 (v/v) with 0.1%formic acid was used for analysis at a flow rate of 40 μL/min, assuggested by the manufacturer (10−100 μL/min). The N2pressure was between 30 and 70 psi and was adjusted manuallywhen necessary to maintain the liquid microjunctionthroughout experiments. Leaking of the solvent from theprobe is avoided through alterations in pressure applied to theESI source but can occur during operation. The voltage appliedto the solvent within the ESI nozzle was 4 kV, and the capillarytemperature of the mass spectrometer was 300 °C.

Tissue Staining. The same tissue sections analyzed byDESI-MSI and adjacent sections analyzed by LMJ-SSP-MSIwere stained using standard H&E staining protocol. Pathologicevaluation was performed using light microscopy.

Lipid and Protein Identification. Lipid species wereidentified using high mass accuracy measurements and higher-energy collision induced dissociation (HCD) tandem MSanalysis, performed on the Q Exactive at 70 000 resolvingpower (m/z 200). Lipid fragmentation patterns were comparedto literature reports and used in conjunction with data from

Analytical Chemistry Article

DOI: 10.1021/acs.analchem.6b02798Anal. Chem. 2016, 88, 11533−11541

11534

Lipidmaps database (www.lipidmaps.org) for identification. Forprotein identification, mass spectra from sequential profilingexperiments with and without the FAIMS separation wereobtained. The data was deconvoluted using the Xtract functionwithin Xcalibur software using a S/N threshold of 3 to obtainmonoisotopic masses of protein species detected duringexperiments. Protein species were analyzed by top-downcollision induced dissociation (CID) tandem MS, performedon a hybrid LTQ-Orbitrap Elite mass spectrometer (ThermoScientific, San Jose, CA) at 120 000 resolving power at m/z200. Protein fragmentation patterns were input into ProSightLite software (http://prosightlite.northwestern.edu/) andcompared with protein amino acid sequences that have beenpreviously observed within the tissue types analyzed. Proteinamino acid sequences were obtained from the UniProtdatabase.2D Imaging Data Analysis. Xcalibur RAW files were

converted into images using FireFly data conversion software(Prosolia, Inc. Indianapolis, IN) and then uploaded into theopen source imaging software packages BioMap (Novartis) orMSiReader33 for visualization. The interpolate function withinBioMap was used to smooth the pixels within the images. Allimages are normalized to the maximum ion intensity within allthe spectra (all pixels) used to create the image.

■ RESULTS AND DISCUSSIONOptimization of FAIMS for Separation of Lipids and

Metabolites. To evaluate FAIMS separation voltages for lipidsand metabolites, a synthetic mixture of standards was preparedand analyzed using ESI and DESI in the negative ion mode.The mixture of lipids contained species from the main classes ofGP commonly detected from biological tissues in the negativeion mode including singly charged PI, PS, PE, PG, and PA,doubly charged CL and gangliosides, three common FA (oleic,lignoceric, and arachidonic), and two representative metabo-lites, n-acetylaspartic acid (NAA) and 2-hydroxyglutarate. ForESI experiments, the synthetic mixture was directly infusedwhile a 2D FAIMS field sweep was performed in which the DFwas stepped in 10 Td increments from 150 to 280 Td with aCV sweep from −1 to +4 Td occurring at every DF value.Increasing DF causes an increase in separation efficiency whiledecreasing total ion current. Thus, an optimal DF was chosenas the field at which lipids and metabolites were most clearlyseparated in the ion chromatogram without detrimental loss oftotal ion current. At this selected DF value, an optimal CF foreach species was chosen as the voltage at which the specific ionhad a maximum total intensity.ESI experiments yielded an optimal DF of 220 Td for the

highest transmittance and separation of lipids and smallmetabolites by FAIMS. At this DF, optimal CFs weredetermined for small metabolites (CF = −0.10 Td), singlycharged GP (CF = 0.99 Td), singly charged FA (CF = 1.17Td), doubly charged gangliosides (CF = 1.80 Td), doublycharged CL (CF = 2.20 Td), and triply charged gangliosides(CF = 2.71 Td; Figure S1). A trend in optimized CF values wasobserved, in which higher CF values were ideal for transmissionof larger, multiply charged ions and lower CF values were moresuitable for the transmission of smaller, singly charged ions.Note that while different optimized CF values allowedseparation of doubly charged CL and gangliosides from singlycharged GP species, separation of different subclasses of singlycharged GP was not clearly observed. DESI-FAIMS optimiza-tion experiments performed on the synthetic lipid and

metabolite mixture yielded similar optimized CF values forthe species of interest at DF of 220 Td. For both ESI and DESI,a drop in total ion current of 1 order of magnitude or more wasobserved with FAIMS separation for all experiments performed.Planar FAIMS devices are known to cause an exponentialdecrease in ion transmission with increasing residence timewithin the device in order to achieve higher separation resolvingpower.34 Thus, despite the short residence time associated withthis chip-based device, this method is not immune to thisdecrease in total ion transmission. Nevertheless, furtherdecrease of noise leads to an overall increase in the S/N forselected molecular ions.34

Optimization of the DESI-FAIMS system was also performedfor imaging lipids and metabolites from a tissue sample in thenegative ion mode. A spot-by-spot DESI analysis of a rat braintissue section was performed using a methodology similar tothe one employed for ESI. At a spatial resolution (step size) of200 μm, the DF was kept constant at the optimized 220 Tdwhile a 1D FAIMS sweep (CF= −1 to 3 Td) was performed(Figure 1A). A total of eight mass spectra, each correspondingto a 0.5 Td CF increment, were acquired at each spot. Usingthis approach, a total of 6 s of analysis time was required foreach spot (pixel). The spectra for each field were extracted andcompiled into 2D ion images, yielding eight sets of 2D DESI-MS images at specific DF and CF values that highlight differentmolecular classes transmitted at the optimized voltages (Figure1B). Although time-consuming, this DESI-FAIMS-MSI experi-ment allowed optimization of separation voltages formetabolites, FA, singly charged GP, doubly charged CL, andgangliosides, directly from a tissue sample. For example,preferential transmission and highest relative abundances ofmetabolites were observed at CF = 0−0.5 Td. At theseparameters, clear spatial distribution of NAA, recently describedas a glioma oncometabolite,35 was observed mainly in the braincortex. In contrast, doubly charged species (CL and ganglio-sides), which are difficult to image within complex biologicaltissue samples due to their low abundances, are nearlyexclusively observed at CF= 2−2.5 Td. We note that a slightdrop in signal intensity was observed with analysis time in eachspot (Figure S2), which could account for lower image qualityfor the species observed at higher CF values. At CF range 2.5−3 Td, little to no ion transmission through the FAIMS devicewas observed. The same optimal ion transmission voltages wereobtained in replicate profiling experiments for the lipid classesinvestigated.

DESI-FAIMS Semiselective Imaging of Cardiolipinsand Gangliosides in Rat Brain Sections. We furtherexplored the DESI-FAIMS methodology for optimized imagingof doubly charged CL and gangliosides from biological tissuesamples. CL are a complex class of GP that exist almostexclusively within the inner mitochondrial membrane of cells.These unique lipids are important structural and functionalcomponents of both normal and diseased tissues, and havebeen increasingly explored in a variety of pathologies.36−38 Wefirst performed DESI-FAIMS-MSI of doubly charged CL withinrat brain tissue sections at the optimized FAIMS parameters inthe negative ion mode. In this static FAIMS approach, both theDF and CF remain constant throughout the entire tissueimaging experiment at the optimized settings (DF = 220 Td,CF = 2.20 Td). Thus, DESI-MSI was operated in the usualrastering mode (0.5 s/pixel), without increase of analysis time,and with a spatial resolution of 200 μm. An evident increase inthe S/N of CL species in comparison to other GP species was

Analytical Chemistry Article

DOI: 10.1021/acs.analchem.6b02798Anal. Chem. 2016, 88, 11533−11541

11535

observed in the mass spectra obtained using DESI-FAIMS-MSIwhen compared to DESI-MSI alone (Figure 2). For example,using DESI-FAIMS-MSI, an increase in the S/N from 116 to203 for an average of 50 spectra was observed for the mostabundant CL species (m/z 749.495, CL (20:4/20:4/30:4/18:1), mass error of −0.93 ppm), with minimal change in signalintensity (from 1.78 × 104 to 1.98 × 104) resulting in imagesthat highlight the CL species over other GP within the tissuesample (Figures 2c and S4). Overall, about 50% increase in theS/N was observed for all CL species detected with the DESI-FAIMS-MSI approach. Most importantly, an increase in thenumber of detectable CL species was observed with theaddition of FAIMS. For example, m/z 797.494, identified asCL(84:18) with a mass error of 1.00 ppm, was only observedby DESI-FAIMS-MSI and undetected with DESI-MSI alone. Intotal, 71 CL species were detected and identified using DESI-FAIMS-MSI, compared to 48 CL species detected when usingDESI-MSI alone, as summarized in Table S1. Acyl chain

composition, but not their positions, was tentatively assignedbased on tandem MS patterns. When tandem MS fragmenta-tion patterns for low abundance CL were unclear, tentativeidentification was based on accurate mass measurements alone.Imaging of doubly charged gangliosides was performed using

serial rat brain tissue sections at optimized FAIMS parameters(DF = 220 Td, CF = 1.8 Td) in the negative ion mode.Gangliosides are a subclass of glycosphingolipids which aremost abundant within the nervous system. Gangliosides haveimportant cellular functions, such as cell signaling and calciumhomeostasis, and have been associated with multiple diseasesincluding Alzheimer’s and Tay-Sachs diseases.39,40 DESI-FAIMS-MSI yielded an increased relative abundance and S/Nof gangliosides compared to DESI-MSI alone. For example, theS/N of the most abundant ganglioside species detected,identified as trisialotetrahexosylganglioside at m/z 1063.530(GT1 d36:1), doubled using FAIMS. Out of the 21 doublycharged ganglioside species detected by DESI-FAIMS-MSI, 7were undetectable by traditional DESI-MSI (Table S2).Gangliosides present at low relative abundances weretentatively identified by accurate mass alone. Note that whileFAIMS allowed improved detection of triply charged ganglio-sides at CF = 2.71 Td using a standard lipid mixture, this chargestate was not observed in biological tissues by DESI-FAIMS orDESI alone due to low abundance of gangliosides at this chargestate under our experimental conditions.

DESI-FAIMS-MSI of Cardiolipins in Human OncocyticThyroid Tumors. We have recently described a diverse groupof CL as molecular markers of oncocytic thyroid tumors,including oxidized CL, adducts of CL and PC, andmonolysocardiolipins.41 However, many of the diagnostic CLare detected at low relative abundances and within close m/zvalues to other interfering GP, which complicates ion isolationfor tandem MS experiments, obscures visualization of their 2Ddistribution, and adds complexity to data analysis. Thus, weapplied the integrated DESI-FAIMS-MSI approach to improveimaging of CL in oncocytic thyroid tumors. Two serial sectionsfrom an oncocytic thyroid tumor were subjected to DESI-MSIunder identical conditions in the negative ion mode, one withstatic mode FAIMS at the optimized CL voltages, and onewithout FAIMS separation. As shown in Figure 3a, DESI-FAIMS-MSI allowed enhanced detection of CLs within theoncocytic thyroid tumor tissue section with reduced interfer-ence from other GP in the m/z 700−800 range. For example,using DESI-MSI alone, PE(O-38:5) at m/z 750.545 had asignal intensity 4 times that of the doubly charged CL(76:9) atm/z 750.502. When FAIMS is applied, CL(76:9) had a signalintensity double that of PE(O-38:5), which facilitates ionisolation, fragmentation, identification, and visualization of its2D distribution. DESI-FAIMS-MSI revealed high relativeabundance of CL within the entire thyroid tissue section, asshown in Figure 3b for m/z 723.479, m/z 737.494, and m/z747.473 identified as CL(18:2/18:2/18:2/18:2 and/or 20:4/18:2/18:2/16:0), CL(20:4/18:2/18:1/18:1, 20:3/18:2/18:2/18:1, 20:2/18:2/18:2/18:2) and CL(22:6/20:4/18:2/16:0),respectively. These results indicate the presence of oncocyticthyroid tumor in the entire tissue section, which was confirmedby histopathologic evaluation of the same H&E stained tissuesection.

LMJ-SSP-FAIMS-MSI of Proteins in Rat Brain TissueSection. We next pursued the integration of LMJ-SSP andFAIMS for biological tissue imaging. LMJ-SSP provides similarmolecular information as DESI from biological tissues with

Figure 1. 2D-FAIMS, spot-by-spot DESI-MS imaging of a rat braintissue section. (a) Schematic of the 2D FAIMS sweep experiment usedto create multiple sets of DESI-MS ion images at varied CF values inone tissue section. Each 200 μm spot on the tissue was analyzed for 6 sfor a total of 8 MS acquisitions, in which a FAIMS sweep occurred,allowing MS acquisition correlated to different CF ranges. (b) DESI-FAIMS-MS ion images for a rat brain tissue section for 6representative lipid and metabolite species, each acquired at DF =220 Td and a different CF range, thus highlighting the increasedtransmission of different molecular species at each CF value. Smallmetabolites transmit at the lowest CF (−1 to +0.5 Td), followed byFA and singly charged complex lipids at midrange CF (0.5−1.5 Td),and doubly charged CL and gangliosides at the high-range CF (1.5−2.5 Td). The images for each ion (vertical columns) are normalized tothe highest ion intensity of that ion, and not across all ions. Scale bar =3 mm.

Analytical Chemistry Article

DOI: 10.1021/acs.analchem.6b02798Anal. Chem. 2016, 88, 11533−11541

11536

higher sensitivity yet lower spatial resolution and image quality,as limited by the probe tip diameter (∼630 μm).42 Metaboliteand lipid optimization experiments performed using LMJ-SSP-FAIMS in the negative ion mode yielded similar results to thoseobtained by DESI-FAIMS. Yet, we found that positive ionmode LMJ-SSP-FAIMS allowed obvious improvements forimaging multiply charged protein species from biological tissuesamples. Thus, we focused our efforts on evaluating andoptimizing LMJ-SSP-FAIMS for protein imaging. Voltageoptimization was performed for midmass range proteins (4−

12 kDa) through LMJ-SSP-FAIMS profiling of a rat brain tissuesection while conducting a 2D-FAIMS sweep from DF 230−280 Td and CF 0−4 Td. Ubiquitin, a well-known protein with amolecular weight of 8.5 kDa, was used for optimization. Thehighest abundance of its 10+ charge state at m/z 857.467 wasfound at DF = 220 Td and CF = 2.55 Td, and thus all staticFAIMS experiments for protein imaging were performed atthese settings.LMJ-SSP profiling experiments of rat brain tissue sections

were performed with and without FAIMS separation in thepositive ion mode under identical experimental conditions. Aclear increase in the relative abundances of multiply chargedprotein peaks was observed in the mass spectra when FAIMSwas applied at optimized voltages for protein transmission(Figure 4a, from m/z 600−1300), despite a 1 order magnitudedrop in total ion current. In total, 84 species attributed toproteins and/or proteoforms were detected using LMJ-SSP-FAIMS, 66 of which were not detected by LMJ-SSP alone. Forexample, the protein isotope pattern for the 9+ ion at m/z821.218 was clearly resolved with a S/N of 28 using LMJ-SSP-FAIMS, while undetectable above a S/N threshold of 3 withLMJ-SSP alone (Figure 4b). Using LMJ-SSP alone, 67 proteinswere detected, 49 of which were not detected by LMJ-SSP-FAIMS. This effect is attributed to lower transmission ofproteins with molecular weight outside of the optimizedFAIMS parameters (4−12 kDa), as well as to the loss of totalsignal intensity associated with FAIMS.34 Out of the six proteinspecies detected by both approaches, four were observed at aslightly higher absolute signal intensity using FAIMS, whileenhancement of S/N was observed for all six of the proteins.For example, a 7-fold increase in S/N (from 23 to 180) wasobserved for the 10+ ion cluster at m/z 857.468 when FAIMSwas applied (Figure 4c). A list of the deconvoluted molecularweights of all protein species detected using LMJ-SSP-FAIMSand LMJ-SSP alone is shown in Figure S6.In an effort to identify detected proteins, top-down CID

tandem MS was performed on protein ions during rat braintissue LMJ-SSP analysis. The 9+ charge state at m/z 857.370(MW = 8559.61 Da) was identified as ubiquitin with a P-scoreof 8.9 × 10−53 (32% sequence coverage, Figure S7, Table S3).The 6+ ions at m/z 828.256 (MW = 4960.48 Da) were

Figure 2. Static DESI-FAIMS-MSI of CL in rat brain tissue. (a) Representative negative ion mode DESI mass spectra acquired without and withFAIMS in the m/z 700−800 range at the optimized doubly charged CL parameters (DF = 220 Td, CF = 2.20 Td), showing a clear increase in therelative abundance of CL species. Without FAIMS and with FAIMS spectra are averages of one DESI line scan, consisting of 68 and 66 scans,respectively. (b) Chart of CL species detected with DESI alone and the DESI-FAIMS integrated approach. (c) 2D DESI-FAIMS ion images forselected CL species (spatial resolution of 200 μm). Voxel versions of the same ion images are shown in Figure S15.

Figure 3. Static DESI-FAIMS-MSI of CLs in an oncocytic thyroidtumor tissue. (a) Representative negative ion mode DESI mass spectraacquired without and with FAIMS in the complex lipid region (m/z700−900) at the optimized doubly charged CL parameters (DF = 220Td, CF = 2.20 Td). Zoom in show spectra from m/z 750−752.Without FAIMS and with FAIMS spectra are averages of a DESI linescan, consisting of 57 and 54 scans, respectively. (b) DESI-FAIMS-MSion images and optical images of the H&E stained oncocytic thyroidtumor sections for three CL species, m/z 723.479, m/z 737.494, andm/z 747.473, showing high relative intensities of CL, a feature ofoncocytic tumors (spatial resolution of 200 μm). Scale bar = 2 mm.

Analytical Chemistry Article

DOI: 10.1021/acs.analchem.6b02798Anal. Chem. 2016, 88, 11533−11541

11537

identified as thymosin β-4 with a P-score of 2.5 × 10−11 (33%sequence coverage, Figure S8, Table S4). Myelin basic protein(MBP) was tentatively attributed to the 20+ ion cluster at m/z706.968 (MW = 14113.23 Da) with a P-score of 3.1 × 10−6

(21% sequence coverage). The cluster of 18+ ions at m/z845.271 (MW = 15187.75 Da) was identified as hemoglobin α-1/2 with a P-score of 9.5 × 10−18 (21% sequence coverage,Figure S9, Table S5). High mass accuracy measurements wereused to confirm protein identity, yielding mass errors below 10ppm for all identified protein ions.LMJ-SSP with and without FAIMS separation was performed

on serial rat brain tissue sections in the positive ion mode.Figure 4 shows 2D ion images of selected protein speciesobtained using LMJ-SSP with a spatial resolution of ∼630 μm.The increase in S/N achieved with LMJ-SSP-FAIMS approachallowed for clear visualization of the spatial distribution of theproteins within the optimized transmission range (Figure 4d).For example, the spatial distribution of the 9+ charge state of anunidentified protein (m/z 939.848, MW = 5449.60 Da) byLMJ-SSP alone was unclear, with signal due to proteindetection and background noise observed throughout thetissue section and the surrounding glass slide. However, theLMJ-SSP-FAIMS image clearly shows protein distribution athigher relative abundances within the white matter of the ratbrain. The same trend was seen for the protein detected at m/z

915.151. LMJ-SSP-FAIMS-MSI also allowed visualization of thespatial distribution of the 6+ charge state of thymosin β-4 (m/z828.256) within the hippocampus of the rat brain (delineatedover optical image of H&E stained tissues), while the 10+charge state of ubiquitin (m/z 857.370) was distributedhomogeneously throughout the tissue section. Proteindistributions obtained by LMJ-SSP-MSI were reproducible (n= 4 for each approach, Figure S13), although small alterationsin experimental parameters including pressure and probedistance to the sample slightly affected ion intensity, whichmost noticeably impacts ion images obtained for ions that arenot localized in a specific region of the tissue. Similar spatialdistributions for these proteins described have been recentlyreported using other ambient ionization MSI techniques.16,31

Proteins with molecular weights outside the optimizedFAIMS voltages are observed at lower absolute intensitieswhen compared to LMJ-SSP alone. For example, MBP has amolecular weight of 15 kDa which is outside the 4−12 kDarange. LMJ-SSP-FAIMS-MSI of its 20+ charge state at m/z706.968 still allows for visualization of its spatial distributiondespite a 2-fold reduction in intensity. Note that lipid ionscould still be detected and imaged using the FAIMS conditionsoptimized from proteins, although at lower relative andabsolute abundances. For example, Figure S14 shows the ionimages obtained from an unwashed rat brain tissue section for

Figure 4. Static LMJ-SSP-FAIMS-MS profiling and imaging of rat brain tissue for protein detection. (a) LMJ-SSP-MS spectra of a rat brain tissuesection, analyzed without FAIMS and with FAIMS optimized for transmission of midmass range proteins (4−12 kDa). Different colored labelsrepresent different charge states of same proteins. (b) Isotope patterns of ubiquitin (z = 10), and (c) an unidentified protein (z = 9), with (red) andwithout (black) FAIMS applied, showing absolute signal intensity and S/N enhancements with the addition of FAIMS. Without FAIMS and withFAIMS mass spectra are averages of a line scan, consisting of 75 scans each. (d) LMJ-SSP-MS and LMJ-SSP-FAIMS-MS ion images of rat braintissue for six representative protein species including ubiquitin, thymosin β-4, MBP, and three unidentified proteins (spatial resolution ∼630 μm).Voxel versions of the same ion images are shown in Figure S15. Scale bar = 3 mm.

Analytical Chemistry Article

DOI: 10.1021/acs.analchem.6b02798Anal. Chem. 2016, 88, 11533−11541

11538

m/z 760.584, PC (34:1), m/z 810.597, PC (38:4), and m/z857.370, ubiquitin, which were detected during the same LMJ-SSP-FAIMS-MSI experiment.LMJ-SSP-FAIMS-MSI of Proteins in Ovarian Cancer

Tissue Sections. The integrated LMJ-SSP-FAIMS approachwas applied to image proteins in human normal ovarian andhigh grade serous ovarian cancer tissue samples. 2D ion imagesrevealed distinct spatial distribution of various protein ionswithin the tissue sections, as shown in Figure 5b for four

selected protein ions: 10+ ion cluster at m/z 857.963 (MW =8559.54 Da), identified as ubiquitin (P-score = 1.6 × 10−20,13% sequence coverage, mass error of 9.35 ppm, Figure S10,Table S6); 5+ ion cluster at m/z 993.501 (MW = 4960.46 Da),identified as thymosin β-4 (P-score = 1.6 × 10−5, 12% sequencecoverage, mass error of 6.15 ppm, Figure S11, Table S7); 16+ion cluster at m/z 946.311 (MW = 15116.76 Da), identified ashemoglobin α-subunit (P-score = 5.3 × 10−11, 26% sequencecoverage, mass error of −7.94 ppm, Figure S12, Table S8); andthe 9+ ion cluster at m/z 1122.038 (MW = 10084.27 Da),proposed as calcyclin (P-score = 5.4 × 10−12, 7% sequencecoverage, mass error of 4.46 ppm). Histopathologic evaluationof H&E stained serial sections revealed the presence of necrotictissue adjacent to the tumor cells in the tumor tissue, while the

normal tissue was homogeneously composed of normal ovariantissue. Comparison of the 2D ion images and the annotatedoptical images of the H&E stained samples revealed thatubiquitin, thymosin β-4, and calcyclin showed higher relativeabundances within the tumor region of the ovarian tumorsample compared with both the necrosis region and the normalovarian tissue. Hemoglobin α-subunit, however, showed higherrelative abundances within the necrosis region compared to theadjacent tumor region and the normal ovarian tissue. The massspectra obtained showed distinct protein distribution for highgrade serous ovarian cancer, necrotic ovarian tissue, and normalovarian tissue under optimized LMJ-SSP-FAIMS conditions(Figure 5a).

■ CONCLUSIONSWe developed and optimized an analytical approach integratingDESI and LMJ-SSP with a chip-based FAIMS device forimaging biological tissue samples. This method allowed forpartial separation of analytes after desorption or extraction frombiological samples, resulting in decreased chemical noise andincreased detection of selected molecular species. Reducinginterferences through FAIMS separation improved the S/N forthe species of interest, aiding in spectra interpretation andimproving ion image quality. Optimization of DF and CFvalues for semitargeted transmission of different molecularclasses can be performed through 2D-FAIMS sweepingexperiments during a profiling or a more time-consumingimaging assay, directly from tissue samples.We show through several examples that this approach is

particularly useful for semiselective imaging of subsets ofmultiply charged lipids and proteins at optimized FAIMSconditions, with no increase in analysis time. For example,improvement in the S/N, ion abundances, and number ofdoubly charged CL and gangliosides species at optimizedFAIMS conditions were observed in rat brain tissue sections.DESI-FAIMS-MSI allowed enhanced detection and imaging ofdiagnostic CL in human oncocytic thyroid tumor with reducedinterference from other GP species, enabling improved ionisolation and identification by tandem MS. LMJ-SSP-FAIMS-MSI was particularly powerful for imaging multiply chargedprotein ions from biological tissue samples. Optimized FAIMSparameters for proteins in the 4−12 kDa range allowed increasein the S/N, number detected, and visualization of the 2D spatialdistribution of 84 protein species within rat brain, even with anoverall decrease in total ion current. LMJ-SSP-FAIMS-MSI ofhuman ovarian tissue samples enabled detection, identification,and correlation of spatial distribution of several proteins withinthe heterogeneous regions of the tissue samples, includingregions of tumor, necrotic, and normal ovarian tissues. This isthe first example of global protein imaging in human cancertissue by ambient ionization MSI.Our results show that FAIMS can be successfully integrated

with DESI and LMJ-SSP for improved detection and imaging ofsubsets of molecular species in biological tissues. Whileseparation using the planar FAIMS device caused a decreasein total ion abundance of 1 order of magnitude or more, anincrease in S/N ratio due to concomitant reduction of noiselevel contributed to improved data quality for MS imagingexperiments.34 Thus, addition of FAIMS to a DESI or LMJ-SSPMSI workflow at optimized conditions is valuable for thedetection and spatial visualization of otherwise undetectablelipids and proteins of diagnostic importance in biologicaltissues. Yet, while FAIMS provides a separation capability to

Figure 5. Static LMJ-SSP-FAIMS-MS profiling and imaging of humannormal and cancerous ovarian tissues. (a) LMJ-SSP-FAIMS-MSspectra of normal ovarian, necrotic, and serous ovarian cancer samplesin which different colored labels represent different charge states ofsame protein species. (b) LMJ-SSP-FAIMS-MS ion images ofubiquitin, thymosin β-4, calcyclin, and hemoglobin α-subunit for anormal ovarian tissue sample compared with the high grade serousovarian tumor sample, containing both necrotic and tumor regions(spatial resolution is ∼630 μm). Optical images of H&E stainedsections show regions of normal ovarian, necrotic, and high gradeserous ovarian tumor. Voxel versions of the same ion images areshown in Figure S15. Scale bar = 2 mm.

Analytical Chemistry Article

DOI: 10.1021/acs.analchem.6b02798Anal. Chem. 2016, 88, 11533−11541

11539

increase data quality in ambient ionization MS imaging, ionsuppression and matrix effects during ionization are stillpertinent and should be considered as these may hinderionization of molecules of interest.43

Further optimization of the integrated system is beingpursued to increase sensitivity and ion transmission. As withother MSI techniques, the extent of molecular informationobtained is significantly less than that achieved with standardHPLC-MS approaches. For example, HPLC-MS proteomicassays of selected tissue regions obtained by laser capturemicrodissection can be performed at similar spatial resolutionas LMJ-SSP, although more time-consuming and laborintensive.44 Evaluation of the molecular coverage achievedwhen compared to standard chromatographic separationtechniques will be sought to better quantify the utility of ourapproach. However, as ambient ionization MSI providesmolecular and spatial information at a fraction of the time,without extensive sample preparation, we expect the integratedapproach described here to be valuable in biomedicalapplications targeted at imaging specific diagnostic lipids andproteins which are otherwise difficult to detect in biologicaltissues.

■ ASSOCIATED CONTENT*S Supporting InformationThe Supporting Information is available free of charge via theInternet at The Supporting Information is available free ofcharge on the ACS Publications website at DOI: 10.1021/acs.analchem.6b02798.

Tables of identified compounds, MS/MS data, ESI-FAIMS optimization data, time-dependent DESI data,CL images for rat brain, summary of detected proteins,LMJ-SSP image replicates, and images illustrating abilityto detect both lipids and proteins. (PDF)

■ AUTHOR INFORMATIONCorresponding Author*E-mail: liviase@utexas.edu.

NotesThe authors declare no competing financial interest.

■ ACKNOWLEDGMENTSThis work was supported by The Welch Foundation (Grant F-1895), the National Inst itutes of Health (grant4R00CA190783-02), and the National Science Foundation(Grant CHE-1559839). We thank Dr. Wendong Yu (Depart-ment of Pathology and Immunology, Baylor College ofMedicine) and Dr. Jinsong Liu (Department of Pathology,The University of Texas MD Anderson Cancer Center) forpathologic analysis. We are grateful to Jialing Zhang, MartaSans Escofet, and John Lin for assistance with experiments anddata analysis. Human tissue samples were provided by theCHTN which is funded by the National Cancer Institute.

■ REFERENCES(1) Dill, A. L.; Eberlin, L. S.; Ifa, D. R.; Cooks, R. G. Chem. Commun.2011, 47, 2741−2746.(2) Fletcher, J. S.; Vickerman, J. C.; Winograd, N. Curr. Opin. Chem.Biol. 2011, 15, 733−740.(3) Seeley, E. H.; Schwamborn, K.; Caprioli, R. M. J. Biol. Chem.2011, 286, 25459−25466.

(4) Wu, C. P.; Dill, A. L.; Eberlin, L. S.; Cooks, R. G.; Ifa, D. R. MassSpectrom. Rev. 2013, 32, 218−243.(5) Hsu, C. C.; Dorrestein, P. C. Curr. Opin. Biotechnol. 2015, 31,24−34.(6) Wiseman, J. M.; Ifa, D. R.; Song, Q. Y.; Cooks, R. G. Angew.Chem., Int. Ed. 2006, 45, 7188−7192.(7) Takats, Z.; Wiseman, J. M.; Gologan, B.; Cooks, R. G. Science2004, 306, 471−473.(8) Ifa, D. R.; Eberlin, L. S. Clin. Chem. 2016, 62, 111−123.(9) Laskin, J.; Heath, B. S.; Roach, P. J.; Cazares, L.; Semmes, O. J.Anal. Chem. 2012, 84, 141−148.(10) Laskin, J.; Lanekoff, I. Anal. Chem. 2016, 88, 52−73.(11) Chen, H. W.; Gamez, G.; Zenobi, R. J. Am. Soc. Mass Spectrom.2009, 20, 1947−1963.(12) Hankin, J. A.; Murphy, R. C. Anal. Chem. 2010, 82, 8476−8484.(13) Yetukuri, L.; Ekroos, K.; Vidal-Puig, A.; Oresic, M. Mol. BioSyst.2008, 4, 121−127.(14) Hsieh, Y.; Casale, R.; Fukuda, E.; Chen, J. W.; Knemeyer, I.;Wingate, J.; Morrison, R.; Korfmacher, W. Rapid Commun. MassSpectrom. 2006, 20, 965−972.(15) Lou, X. W.; van Dongen, J. L. J.; Vekemans, J.; Meijer, E. W.Rapid Commun. Mass Spectrom. 2009, 23, 3077−3082.(16) Hsu, C. C.; Chou, P. T.; Zare, R. N. Anal. Chem. 2015, 87,11171−11175.(17) Kertesz, V.; Weiskittel, T. M.; Van Berkel, G. J. Anal. Bioanal.Chem. 2015, 407, 2117−2125.(18) Lombard-Banek, C.; Moody, S. A.; Nemes, P. Angew. Chem., Int.Ed. 2016, 55, 2454−2458.(19) Collins, D. C.; Lee, M. L. Anal. Bioanal. Chem. 2002, 372, 66−73.(20) Kanu, A. B.; Dwivedi, P.; Tam, M.; Matz, L.; Hill, H. H. J. MassSpectrom. 2008, 43, 1−22.(21) Ray, J. A.; Kushnir, M. M.; Yost, R. A.; Rockwood, A. L.; Meikle,A. W. Clin. Chim. Acta 2015, 438, 330−336.(22) Guevremont, R. Journal of Chromatography A 2004, 1058, 3−19.(23) Kolakowski, B. M.; Mester, Z. Analyst 2007, 132, 842−864.(24) Da Costa, C.; Turner, M.; Reynolds, J. C.; Whitmarsh, S.;Lynch, T.; Creaser, C. S. Anal. Chem. 2016, 88, 2453−2458.(25) Manicke, N. E.; Belford, M. J. Am. Soc. Mass Spectrom. 2015, 26,701−705.(26) Griffiths, R. L.; Dexter, A.; Creese, A. J.; Cooper, H. J. Analyst2015, 140, 6879−6885.(27) Li, H.; Smith, B. K.; Mark, L.; Nemes, P.; Nazarian, J.; Vertes, A.Int. J. Mass Spectrom. 2015, 377, 681−689.(28) Roscioli, K. M.; Tufariello, J. A.; Zhang, X.; Li, S. X.; Goetz, G.H.; Cheng, G. L.; Siems, W. F.; Hill, H. H. Analyst 2014, 139, 1740−1750.(29) Galhena, A. S.; Harris, G. A.; Kwasnik, M.; Fernandez, F. M.Anal. Chem. 2010, 82, 9159−9163.(30) Sarsby, J.; Griffiths, R. L.; Race, A. M.; Bunch, J.; Randall, E. C.;Creese, A. J.; Cooper, H. J. Anal. Chem. 2015, 87, 6794−6800.(31) Griffiths, R. L.; Creese, A. J.; Race, A. M.; Bunch, J.; Cooper, H.J. Anal. Chem. 2016, 88, 6758−6766.(32) Eberlin, L. S.; Ferreira, C. R.; Dill, A. L.; Ifa, D. R.; Cheng, L.;Cooks, R. G. ChemBioChem 2011, 12, 2129−2132.(33) Robichaud, G.; Garrard, K. P.; Barry, J. A.; Muddiman, D. C. J.Am. Soc. Mass Spectrom. 2013, 24, 718−721.(34) Swearingen, K. E.; Moritz, R. L. Expert Rev. Proteomics 2012, 9,505−517.(35) Jarmusch, A. K.; Pirro, V.; Baird, Z.; Hattab, E. M.; Cohen-Gadol, A. A.; Cooks, R. G. Proc. Natl. Acad. Sci. U. S. A. 2016, 113,1486−1491.(36) Ji, J.; Kline, A. E.; Amoscato, A.; Samhan-Arias, A. K.; Sparvero,L. J.; Tyurin, V. A.; Tyurina, Y. Y.; Fink, B.; Manole, M. D.; Puccio, A.M.; Okonkwo, D. O.; Cheng, J. P.; Alexander, H.; Clark, R. S. B.;Kochanek, P. M.; Wipf, P.; Kagan, V. E.; Bayir, H. Nat. Neurosci. 2012,15, 1407−1413.

Analytical Chemistry Article

DOI: 10.1021/acs.analchem.6b02798Anal. Chem. 2016, 88, 11533−11541

11540

(37) Sapandowski, A.; Stope, M.; Evert, K.; Evert, M.; Zimmermann,U.; Peter, D.; Paege, I.; Burchardt, M.; Schild, L. Mol. Cell. Biochem.2015, 410, 175−185.(38) Corcelli, A.; Angelini, R.; Lobasso, S.; Bowron, A.; Steward, C.FASEB J. 2015, 29.(39) Sandhoff, K.; Harzer, K. J. Neurosci. 2013, 33, 10195−10208.(40) Yu, R. K.; Tsai, Y. T.; Ariga, T. Neurochem. Res. 2012, 37, 1230−1244.(41) Zhang, J.; Yu, W.; Ryu, S.; Lin, J.; Buentello, G.; Tibshirani, R.;Suliburk, J. W.; Eberlin, L. S. Cancer Res. 2016, DOI: 10.1158/0008-5472.CAN-16-1545.(42) Prideaux, B.; ElNaggar, M. S.; Zimmerman, M.; Wiseman, J. M.;Li, X.; Dartois, V. Int. J. Mass Spectrom. 2015, 377, 699−708.(43) Chen, H.; Gamez, G.; Zenobi, R. J. Am. Soc. Mass Spectrom.2009, 20, 1947−1963.(44) Zang, L.; Toy, D. P.; Hancock, W. S.; Sgroi, D. C.; Karger, B. L.J. Proteome Res. 2004, 3, 604−612.

Analytical Chemistry Article

DOI: 10.1021/acs.analchem.6b02798Anal. Chem. 2016, 88, 11533−11541

11541